Consistent correction of census and vital for Thailand, 1960-80 .-

PAPERS OF THE EAST-WEST POPULATION INSTITUTE published about eighttimes a year facilitate early dissemination of research findings and policyshyrelevant reports on tile populations of Asia the Pacific and the United StatesCost per copy $3 NOTE TO CONTRIBUTORS The Population Institute considers unsoicited aswell as commissioned manuscripts for the Paper Series Appropriate topics aredemographic trends and estimation fertility and tamijy structure economicdevelopment and human resources urbanization and migration and populashytion policies and programs All manuscripts are reviewed In selectingmanus-ripts for publication the Institute considers quality of scholarship anduseful iess to public officials and other professionals in the fieid of populationt also seeks contributions reflecting diverse cultural and disciplinary perspecshytives on population The series c-m accomrnodate articles not necessarily suitedfor journals because of unusual length or treatment of subject All copy mustbe typed double-spaced Foi additional information oit manuscript preparashytion write to Editor Paper Series

OTHER SERIAL PUBLICATIONS OF THE EAST-WEST POPULATION INSTITUTE Working Papers are circulated for comment and to infoirm interested colleaguesabout work in progress at the East-West Population In1itute They are intendedto complement evidence of completed work as reflected in Papers of the East-West Population Institute and the Reprint Series $1 per copy Reprint Seri 2s brings selected articles originating from Institute research butpublished elsewhee to the attention of population specialists who might nototherwise see them Single copies available upon request Asian and Pacific Population Forum brings articles of potential value in policyformulation trogram administration and research to the notice of policymakshyers piofessionals and scholars concerned with populatioc matters in the Asia-Pacific region Published quarterly in November February May and AugustAnnual subscription rate $12 (free to individuals and organizations involvedin popuJation-related ork) Serial publications except Working Papers aie available without charge tolibraries serving populaion specialsts and to pofessionals and scholars in thefield of population Requests describing the nature of the research or programand the intended use of the publications should be addressed to PublicationSales and Distribution East-West Center

Eia-t-West Population InstituteEast-West Center 1777 East-West Road

Director Lee-lay ChoHonolulu Hawaii 96848 Editor David EllisSenior Editor Sandra E Ward

Consistent correction of census

and vital registration data for Thailand 1960-80

Norman Y Luther Neramit Dhanasakdi

and Fred Arnold

Number 103 December 1986

PAPERS OF THE EAST-WEST POPULATION II ISTITUTE

NOR AN Y LUTHER is Associate Professor Department of Pure and Ap plied Mathematics Washingtu State University and Research Fellow East-West Population Institute NERAMIT DHANASAKDI is Chief of the Popu lation and Housing Unit Population Statistics Branch Population Surve) Division National Statistical Office Bangkok FRED ARNOLD is a Researd-Associate EastWes Population Institute

Library of Congress Cataloging-in-Publication Data

Luther Norman Y 1936-Consistent correction of census and vital registrashy

tion data for Thailand 1960-80

(Papers of the East-West Population Institute 103) Bibliography p 1 Registers of births etc-Thailand

2 Thailand-Census 1960 3 Thailand-Census 1970 4 Thailand-Census 1980 5 Thailand-Statistical services 1 Nerarnit Dhanasakdi l46- I1 Arnold Fred 11lTdEi IV Series HA39T4L87 1986 30460723 86-32807 ISBN 0-86638-089-2

CONTENTS

Acknowledgments vii

Births 23

Deaths

Abstract 1

Method 2

Application to Thailand 1960-80 3

Principal findings 6

Censuses 6

Vital registration 9

Life tables 12

Summary and conclusion 19

Appendix A Derivation of the preliminary correction factors 23

Censuses 24 25

Appendix B Tables 29

Appendix C Formulas for deriving the life tables 35

References 37

V

FIGURES AND TABLES

Figures 1 Preliminary and final correction factors for males Thailand censuses

of 1960 1970 and 1980 11

2 Preliminary and final correction factors for females Thailand censuses of 1960 1970 and 1980 12

3 Preliminary and final correction factors for intercensal registered deaths for males Thailand 1960-65 1965-70 1970-75 and 1975-80 17

4 Preliminary and final correction factors for intercensal registered deaths for females Thailand 1960-65 1965-70 1970-75 and 1975-80 18

5 Corrected life table probabilities of dying rq Thailand 1960-65 and 1975-80 19

Tables

1 Reported and final corrected census age distributions and estimates of completeness for nales Thailand censuses of 1960 1970 and 1980 7

2 Reported and final corrected census age distributions and estimates of completeness for females Thailand censuses of 1960 1970 and 1980 8

3 Final correction factors Thailand censuses of 1960 1970 and 1980 9

4 Completeness estimates (percentages) Thailand censuses of 1960 1970 and 1980 10)

5 Registered and final corrected intercensal births and deaths and estishymates of registration completeness for males Thailand 1960-05 1965-70 1970-75 and 1975-80 13

6 Registered and final corrected intercensal births and deaths and estishymates of registration completeness for females Thailand 1960-65 1965-70 1970-7 and 1975-80 14

7 Final correction factors for registered births and deaths Thailand 1960-65 1965-70 1970-75 ald 1975-80 15

8 Completeness estimates (percentages) for death registration Thailand 1960-65 1965-70 1970-75 and t975-80 16

9 Derived intercensal life tables for males Thailand 1960-65 1965-70 1970-75 and 1975-80 20

10 Derived intercensal life tables for females Thailand 1960-65 1965-70 1970-75 and 1t175-80) 20

vi

11 Estimated midcensus age distributions for males and females Thailand 1965 and 1975 (1 April) 22

B 1 Preliminarily corrected census age-sex distributions Thailand 1960 1970 and 1980 30

B2 Preliminary correction factors Thailand censuses of 1960 1970 and 1980 31

B3 Coale-Demeny model North life tables for determining preliminarycorrection factors for registered deaths for males Thailand 1960-65 1965-70 1970-75 and 1975-80 32

B4 Coale-Demeny model North life tables for determining preliminary correction factors for registered deaths for females Thailand 1960-65 1965-70 1970-75 and 1975-80 33

B5 Preliminary correction factors for registered births and deaths Thaishyland 1960-65 1965-70 1970-75 and 1975-80 34

vii

ACKNOWLEDGMENTS

We are grateful to Wiwit Siripak and Chintana Pejaranonda of the National Statistical Office Bangkok to Robert D Retherford of the East-West Popushylation Institute for helpful discussions and comments and to Robin Loomis of the East-West Population Institute for research assistance Support for this research project was provided by the Office of Population US Agency for International Development under a cooperative agreement with the East-West Population Institute

ABSTRACT A new procedure for the simultaneous and consistent corretshytion of two or more censuses and intercensal registered births and deaths is applied to Thailands censuses of 1960 1970 and 1980 and to registered births and deaths for tile period 1960-80 The procedure begins with a set of prelimishynarv correction fctors that are inconsistent because they are derived from a variety of eisting sources File procedure then identifies the set of consisshytent correction factors that are closest to the preliminary factors

The results show marked tinderenuneration fithe 0-4 age group in each Census especially the 1Q0N) Censtl in which age rounding was apparently prevalent Age eaggeration at tihe oldest ages is also iidicated at least in the I)70t and lc80 censuses I3irth registration completeness rose during 100-80 from about 82 to 85 percent lfo males oIdL from about 78 to S5 percent for females NeVtrtlilesS death re istraitin ompleent for both sexes hovered within the t] to 07 pertent ranige declining slightly until the early ly7t)s when it leveled oft Mrbegainll rti atge grolup )-4 had the greatest uIinderregisshytraltion of deatllh and tnderreistratiioin at this age ipparentlly increased tllroLgll)lt the period AlthoughWl lite tpctaIcv increased fol botlh durshysexes ilg tihe perieid temale uperioiitv in life epectaincv alsoIncreased The avershydge MIlnber of Veas Itift expectaicy i esti mated tilihdvt ilcredsd from

to in to-2 fir femllalesQ3 for feniahegt and 5-18 males I)l)t- - h and 581 for mal in 11)7 80

The lha ilaind censuses Of I)00 1970 and 1980 and vital registration during the period N0t)-80) have betn the subject of variolS etimate of ensus tinshydercounlt tdiusted census e-se distributions underregistratiolmof births and datllhs intercensal life tables and other demographic variables many OttwhichI 0re men tit md beliOw I OWtr those estimaltes as a whoile conshytain nynV incolsistncies lhe ptllpos of this paper is to prOvide a neW CI sistetl set f estinai or thes data making sflective use of the previ-OLts tstillIates

Il accomplish thli is pu rp Iso a new pit(icedILre to] the siniultlnetois and Coilsistlt cirloctioll if to or ire tcensu-es and intercensal registered birlths and deaths (Ilther and Ntt hrford It5(a It)Sfb) is appleHd to Thailtnds ceInsuses of I0t)t) 197(0 and 190 nild to the registered births and deaths f ilfh intrveililrg period Of 1t0t0-8( 5pecificallv for each se the procedure provides estinmtes Of (I) cortctioi fctors for intrcensl registered births (2) age-specific correction iactors for ii terce nsal registered deaths (3) age-specific correction factors ftor the age distributii at each CensuS and (4) life tables fol the intr1censl[ pe t tiis such that the corrected registered births anI deaths the cirreedCte l SlS age distributions and the estimated life tables are mutually consistent Furthermore the correctshyId births satisfx - sex ratio Consistencv here leans that the usualspecified intercensal dei ilraplliC balancing equtiaons relating censIs age distribushy

2 Consistent Correction of Thai Data

tions births and deaths by age are exactly satisfied Net international migrashytion was negligible for the period 1960-80 so that it may be ignored Tile volume of permanent immigration and emigration was small throughout this period Temporary migration of contract workers to the Middle East began to increase in 1978 but it did not become substantial until after the 1980 census

METHOD The procedure begins with a set of preliminary correction factors that are not necessarily consistent These preliminary correction factors are derived from previous estimates and by eisting methods of demographic analyshysis as detailed below Starting from this set of preliminary correction facshytors and the demographic data themselves the new procedure specifies a best or Optimal set of final conisistent correction factors that is closest to the set of preliminary correction factors The optimization procedure finds this closest set by Use Of the pri ncipie from the mathematics Of finiteshydimensional vector spaces that there exists in a hyperplane a unique point of iinimmi in dista nce from a fixed Poi nt n0t in tiht hperpla ne The met hshyodologyv is explained in I tther and Retherford ( lt)6a 1986b) tte latter source gives details for the ntn-mathematicallv-inclined reader It is shown in those sources that 1y utilizi nv the concept (f weighted distance thle procedure allows the analyst to weight the preliminary est imattes dccordshying to the analysts asessmnt Otftheir ikcuracY 1e rise of proportional weights as described in the applicatitti n tOfl roCedCrime tt Sot htIKoreaI data for 1970-80 (Luther aMid Retherftord 98ha IL)Sjbb) will spread the COrshyrection evenly over the prelim inalrV correction tacttrs This is also approprishyate for the Thailand application since we would assess all Of the preliminary correction factors derived below to be of comparable accuraCV

The three-census application to Ihailand given in this paper illustrates ami important advantage of this procedure that it may be used to correct consistently more than two censuses at a time along wth intercensa births and deaths In contrast most previous procedures correct just two ciisLses at a time (eg Demenv and Shorter 1968 P reston 1983) and that typically leads to inconsistent results for the intermediate censu ses from the separate applications of the procedure In particular most previotIs analyses of Thailand data have involved just two successive ceIsuIses (eg Fulton 1979 Arnold and Phananiramai 1t75 iPejaranonda Arnold and I lauser 1983)

Another advntage of the consistent correction procedure is that it alshylows the researcher great leeway in the aniount and type of structure that is imposed For example one may reluire the final age-specific correction factors to be the same from census to census as with the Deneny-Shorter technique (Denieny and Shorter 1968) and the variable methods (Preston

3 Application to Thailand 1960-80

1983) No such assumptions are made for the Thai application given here however This latter more flexible approach which is often more realistic and appropriate was adopted for the Korean application as well (Lutherand Retherford 1986a 1986b) Or one may compromise and assume that the final correction factors follow some trend As an example of a different type of restriction that the procedure allows we have fixed the sex ratio for the corrected Thai birth In fact the only conditiolis that must be satisshyfied by the restrictions involving the correction tactors are that these reshystrictions be linear in the factors and hold when all factors are zero

The proced ore however has limitations The primary one is that the final correction factr are snsiIive to errors in the prelininary correction factors ConseI 1uentlV it is inpo0rt ant to estimate the preliminary correcshytion factors as accurately as possible

In alnV case the final correcti~m factor given by the procedure yieldconsistent adjUtstments of the data The degree totwhich these adjustments reactuall corrctiIns depends on the accuracy of the preliminary correcshytion tacdrs

ITI ICAIION TO TI IAILAND 1960-80 All censuses are treated Isif they occurred on I April although the 1960 census wa taken iio 2April [Ie ft or five-Vear intercensal periods that will be tIjslentl V nent iimed are tOSte separated by I April of the yearsIL)(() 10)b IL)7() and 1()1)75 8

The input into thiprocedtre and specifically into theCmLpter proshygram (which may be ohbtaintd trOil the author) consists o f (1)the 1)6011)70 and I )8)reported ceisos ageC distribo tioIs by se and five-year age griops (open-ended at age 70) given in bibles I and 2 in the next section with each censui ag distribution unilormly preolultiplied by a correction factor based tila Lentsos underenlneratioll estimate (the details of this premtltiplication are given with the derivation of the preliminary correcshytion lcttirs inAppenidix A) (21 tttal registered interce nsalI births by sex for each Of tile four five-year inltrcensaI periods given in bibles 5 and 6 (3) total registered intercensal deaths by sex a11d five-yearage groups (openshyended at age 70) h ir each if tile four five-year initercerisal perio ds againgiven in bibles 5 and 6 (0) 1 I and J for each of the fourVl ues by sex five-year intercensal periods obtained from Coale-)enieny model North life tables (levels 16825 17 175 17225 and 17875 respectively for males given inAppendix lable B3 and 170 176 182 and 19) respectively for fenales given in Appendix ible BA) in order to yield alt estimate of the separation factor for deaths at ages 0-4 by sex for each of the four five-year intercensal periods (3)a specified sex ratio at birth of I058 for each of the four fiveshyyear intercensal -eriods and (6) estimated preliminary correction factors

4 Consistent Correction of Thai Data

for each of the demographic quantities listed in (1) (2) and (3) Because there are fifteen age groups (fourteen five-year age groups plus the openshyended one) for each sex there wil be forty-five census correction factors (fifteen for each census) and sixtv c eath correction factor (fifteen for each five-year intercensal period) If we add to this the four correction factors for births for each sex (one for each five-year intercensal perk d) we have a total of 109 correction factors for each sex The preliminary correction factors are put into the computer program with male factors first and feshymale factors following The ordilr of preliminary correction factors for each sex is births first (in chronological order) then censuses (youngest to oldshyest age group in first census in second census etc) then deaths (younshygest to oldest age group in first intercensal period in second intercensal period etc)

Because registered births and deaths (by age) for Thailand are tabushylated by calendar year it was necessary to interpolate to obtain registeredbirths and deaths for the four five-ytar intercensal periods For exampleregistered births for the first five-year intercensal period 1 April 1960 to 1 April 1965 include three-fourths of the births for 1960 all of the births for 1961-64 and one-fourth of the births fo 1965 The same interpolation procedure was used for deaths by age group

Coale-[Demenv model North life tables have been used frequently in the analysis of thai data for example by Pejaranonda Arnold and Choe (1985) and by Fulton (1979) after a careful study comparing the fit of the different Coale-Demenv (1966) model life table families to Thai mortalityexperience We opted for such model tables rather than those resulting from the Surveys of lopulatiOn Change (SPC) of 1964-07 and 1974-76 (ThailandNational Statistical Office 1969 1978) because of the irregularity of some SPC qjvalues probably due to the small sample size At the mnortalitv levshyels listed in (4) above the (Coale-Denmeny model North tables are appliedboth for estimating separation factors for deaths at ages 0-4 and for derivshying the preliminary correction factors for deaths The one exception involves the later adult ages where it Was judged that the SIPC life tables more acshycurately reflect Thai mortality experience during 1960-8() therefore the SPC life tables were employed for both sexes rather than the model North life tables in order to derive the preliminary correction factors for deaths for the 70+ age group for each Of the four five-year intercensal periods

Moreover the life expectancy values (e)from the SPIC life tables detershymined the levels of the model North life tables that were used For examshypie the 1964-67 and 1974-76 SPC male life tables list Cvalues of 5627 and

Application to Thailand 1960-80 5

5800 respectively These values were assumed for the dates 1 April 1965 and I April 1975 which are encompassed by the respective SPCs Model North I k of 170 and 177 were matched to these male c values at these s accordingly life tables for males for each Of the four five-year intershycensal periods were obtained by linear interpolation taking the reference dates to be the midpoints of the five-year periods

The same process was used to obtain a female lite table for each fiveshy1)year intercensal period except that level was used for the period 1April

1975 to 1 April 198i) rather than 188 as would result from linear interpolashytion The decision to use this higher level was based on evidence that life expectancy for females was accelerating in the 1)70s at least in relation to that for males (Ifill 197Q) MoreoVer the results from using the lower figure of 188 contradicted preliminary evidelce from the 1984-86 SPC that death registratin completeness leveled off in the i7(s rather than continuing its earlier decline

The preliminary correction factors for deaths v sex and by five-year intercensal peiod for the 70 1 age group were obtdined from the SPC life tables as follows The SPClife tables for 19614-67 and 1974-76 were ascribed to the dates I April 1N65 and 1 April 1975 respectiveJly Linear interpolation was performed and again the reference dates were taken to be the midshypoints of the five-year intercen sal intervals

The estimate of 1058 as the true sex ratit at birth was based on the average of the values 1055 and 1062 obtained by the 1964-67 and 1975-76SPCs The same value of 1058 was input for each of the five-year intercenshy

sal periods since there is no reasol to expect the sex ratio at birth to vary from one five-year period to the next The correction procedure then reshyqIuires that the corrected registered births satisfy the specified sex ratio of 1058 for each five-yea r ilterensial period This featuore has been added to the correction prcedu re ince its origilnal description in Iuther and Rethershyfoird ( 19 80a 198b)

The complete details Of the derivation of the preliminary correction facshytors art given in Appendix The reader who is interested in the nechanshyics as well as the results of the cnsistent correction proced ure can refer to that appendix As ment ineCL before the more theoretical aspects of the methodolhgy of the prwedlre are g yV lilh r anl Retherford (19 86ain ill 1986b)

Finally it vs w rth noting that the cost 0f running the computer proshygram for this application involving three Thai censuses aId tour five-year intercensal periods is only about $16

6 Consistent Correction of Thai Data

PRINCIPAL FINDINGS The principal findings are given in Tables 1- H Figures 1-5 and Appendix Tables B1-B5

Censuses Tables 1-4 and Figures 1 and 2 present the final results for the censuses In Figures 1and 2 the final correction factors of Table 3 are compared with the corresponding preliminary correction factors given in Appendix Table B2

Especially striking is the high degree of underenumeration of the 0-4 age group for both sexes for all censuses but especially for the 1960 census (Table 4) This is consistent with the observationof Clhamratrithirorg Deshybavalya and Knodel (1978) Panel on Thailand ()1980) United Nations (1966)and others that the 0-4 age group was especially underenunmerated in the 1960 census because of age rounding owing to the fact that the census quesshytion asked for-ages of household members rather than dates of births Thus the ages of children between exact ages 4 and 5 were frequently rounded upward to age 5 This was less likely to occur in the 1070) and 1980 censhysuses in which date of birth was asked rather than age

If there was rounding to the nearest age at all ages in the 1960 census then the count ot say the 5-9 age group (exact ages 5 to 1t) included some children between exact ages 41 and 5 and excluded some between exact ages 91 and 10 Suppose the extent of rounding was substantial and of similar proportions througlhoJt the entire population Since the total popushylation between exact ages 4 and 5 exceeds that between exact ages 92 and 10 such rounding woud tend to result in a relative overcount of tle 5-9 age group and for the same reason those age groups ibove it Thus for those age spans in which enumeration was relatively complete slight overenumeration would appeir The results show this to be true for ages20-49 for both males and females in the 1960 census (see Table 4) Since the 1960 census is belieVed to have given a relatively complete count for ages 5 and above (United Nations 196) -AucIh slight overellumel ation for some age groups is plausible

On the other hand the results imply that the 197(0 census contained considerable undereluileration at ages 20-41) as well as t ages 90-69 Of course the total p pulation fthe 1970 census was substantialh more unshyderenumerated thln thel1961) census (se eg Fulton 1979 Preston and Hill 1980 and IlutherI1983 as well as Thble 4)

Table 4 also indicates considerabie age exaeratiOin at the older agesin both theI1970 and I1980 censuses althilgh the age exaggeration for males in1970 was not so fficient to fftset the gentrla tLIndercoul t One cannot disshycern from the procedue however wChet her such age exaggeration occurred